Method for translating patent documents

Abstract

To provide a method for translating a patent document capable of more efficiently preparing an accurate translated sentence at the time of translating the patent document by using machine translating.SOLUTION: A reference text providing step of providing, to a natural language processing artificial intelligence having at least a translation function and a dialogue function, an original text including "claims" and "specification" described in a first language as a reference text for performing machine translation, and a translated text of "claims" obtained by translating "claims" into a second language; The method for translating a patent document according to claim 1, further comprising: adding a condition for translating the "specification" into the second language based on the "claim" and the translated text of the "claim"; and generating a translated text of the "specification" by translating the "specification" into the second language according to the condition.SELECTED DRAWING: None

Description

[Technical Field]

[0001] The present invention relates to a method for translating patent documents. [Background technology]

[0002] With the recent advances in machine translation technology, machine translation is becoming increasingly popular when translators translate various documents, such as patent documents, with the aim of improving the efficiency of translation work. However, simply translating source text directly into a machine translation can result in various problems (translation errors), such as mistranslations, missing parts, or the source text being translated in an unnatural style or word order. For this reason, to improve the quality of translations obtained using machine translation, translators and checkers who specialize in checking and correcting translations may (i) pre-edit the source text before machine translation to create pre-edited source text, which is then used for machine translation, and / or (ii) post-edit the machine-translated text obtained by machine translation to create a post-edited translation.

[0003] Regarding the process of creating a translation using such machine translation, for example, Patent Document 1 proposes a technology that reduces the burden of post-editing a text that has been machine-translated using neural machine translation. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2022-57482 Summary of the Invention [Problem to be solved by the invention]

[0005] In patent documents, for example, when filing a second patent application claiming priority based on a first patent application written in a first language, it is necessary to prepare a translation of the specification and other documents of the first patent application into the official language (second language) of the second country. Accuracy is required for such translations. However, despite recent improvements in the accuracy of neural machine translation, the nature of patent documents means that simply translating the target text is often insufficient. Furthermore, improving the accuracy of machine-translated text requires proofreading, such as post-editing. However, proofreading a machine-translated text requires a significant amount of work.

[0006] The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a method for translating patent documents that can more efficiently create accurate translations when translating patent documents using machine translation. [Means for solving the problem]

[0007] The above object is achieved by the present invention as follows: The patent document translation method of the present invention is a method for machine-translating a "specification" into a second language using original text including "claims" and a "description" written in a first language and a translation of the "claims" obtained by translating the "claims" into the second language, the method comprising the steps of: a reference text assignment step of assigning the translation of the "claims" and the original text as reference text when performing machine translation to a natural language processing artificial intelligence (NLP) residing in a terminal computer or on a network system connected to the terminal computer and having at least a translation function and an interactive function; a translation condition assignment step of assigning translation conditions to the NLP for translating the "description" into the second language; and a translation step of generating a translation of the "description" by the NLP in accordance with the translation conditions.

[0008] It is preferable that one embodiment of the patent document translation method of the present invention further includes a step of extracting potential typographical errors, in which the natural language processing artificial intelligence compares specific wording in the translated "Claims" with the entire translated "Specification" to extract wording from the translated "Specification" that is not completely identical to the specific wording but is highly similar to the specific wording.

[0009] In another embodiment of the patent document translation method of the present invention, the translation conditions preferably include at least one of the following conditions (a) to (c): (a) When translating the "specification" into a second language, the translation shall conform to the wording of the translation of the "claims." (b) When translating the "Description" into a second language, translate the "Description" word for word. (c) Each word in the text of the "Specification" must be translated into a second language using only one translation.

[0010] In another embodiment of the patent document translation method of the present invention, it is preferable that the first language is any one language selected from the group consisting of Japanese, English, Chinese, and Korean, and the second language is selected from the group consisting of Japanese, English, Chinese, and Korean, and is a language different from the first language.

[0011] In another embodiment of the patent document translation method of the present invention, it is preferable that the first language is Japanese and the second language is English.

[0012] In another embodiment of the patent document translation method of the present invention, it is preferable that the first language is English and the second language is Japanese.

[0013] In another embodiment of the patent document translation method of the present invention, it is preferable that the first language is Japanese and that at least one claim included in the "Claims" written in the first language has a character count of 300 characters or more.

[0014] In another embodiment of the patent document translation method of the present invention, the natural language processing artificial intelligence is preferably any one selected from the group consisting of ChatGPT, Google Assistant, Microsoft Cortana, IBM Watson Assistant, and Baidu DuerOS.

[0015] In another embodiment of the patent document translation method of the present invention, the machine translation process is preferably performed in part or in its entirety using robotic process automation. [Effects of the Invention]

[0016] According to the present invention, it is possible to provide a method for translating a patent document that can more efficiently create an accurate translation when translating a patent document using machine translation. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a schematic diagram showing an example of a computer terminal and a network system used in the patent document translation method of the present invention. [Figure 2] FIG. 1 is a schematic diagram showing an example of a user interface of a natural language processing artificial intelligence used in the patent document translation method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0018] <<Natural Language Processing Artificial Intelligence>> The patent document translation method of this embodiment utilizes a natural language processing artificial intelligence (NLP) equipped with at least a translation function and a dialogue function. Therefore, before describing the patent document translation method of this embodiment, we will first explain the overview, fundamental problems, and usage of NLP. In the following description, "artificial intelligence" may be abbreviated as "AI."

[0019] In recent years, with the development of artificial intelligence (AI) technology specialized in specific fields (so-called narrow AI), AI that understands and generates text (natural language processing AI) is becoming widely used as a type of narrow AI. Natural language processing AI is AI that understands and generates human language, and functionally it is mainly classified into six types shown in items (1) to (6) below. In addition to function-specific AI that specializes in only specific functions and has been widely used for more than several years, there are also highly versatile general-purpose AI that simultaneously combine multiple functions and have become rapidly popular in the past one to two years (for example, ChatGPT, which has the functions of items (1) to (4) and (6) below).

[0020] (1) Machine translation (neural machine translation): AI that translates text between different languages ​​(e.g., commercial services such as Google Translate and DeepL). (2) Text summarization: AI that summarizes long texts in a short form (for example, automatic summarization of news articles). (3) Sentiment analysis: AI that extracts emotions and thoughts from text (for example, analyzing social media comments). (4) Dialogue: AI that generates appropriate answers to user questions and carries out user commands and requests (e.g., FAQ chatbots). (5) Speech recognition and generation: AI that converts voice data into text and reads text aloud (for example, voice assistants, text-to-speech software, etc.). (6) Grammar Check: AI that detects and corrects grammar and spelling errors in text (e.g., commercial services such as Grammarly).

[0021] Machine translation performed in natural language processing AI is called neural machine translation, which uses deep learning technology to learn language patterns from large amounts of diverse text data and then uses the learning results to perform translation. Therefore, if the amount and diversity of text data available for training is limited, the accuracy of machine translation will decrease. Furthermore, neural machine translation uses mechanisms and algorithms different from those of the human brain (e.g., translator) to perform translation, and deep learning requires large amounts of diverse text data, making it inevitable that inappropriate or low-quality data will be mixed in with this text data. Furthermore, the text data prepared for training contains sentences of varying lengths and structural complexity. The longer the sentences and the more complex their structures are compared to sentences of average length and structural complexity, the fewer samples there will be and the less diverse they will be. This means that compared to sentences of average length and structural complexity, longer sentences with more complex structures have less training data and less diversity (comparing AI to humans, it would be like it is understudied), making it more difficult for natural language processing AI to accurately understand and generate (in the case of machine translation, translation errors are more likely to occur).

[0022] Considering the principles of machine translation performed by natural language processing AI described above and the problems that inevitably arise with it, it can be said that there will inevitably be a discrepancy in the content and quality of translations produced by a translator with appropriate translation skills and machine-translated text, and this tendency becomes particularly pronounced the longer the sentence and the more complex its structure. Therefore, even if future technological developments improve the translation accuracy of neural machine translation, it is expected that some form of editing and proofreading, such as post-editing, will remain necessary in the translation process using machine translation for the foreseeable future when a certain level of translation quality is required.

[0023] The natural language processing AI used in the patent document translation method of this embodiment has at least a translation function and a dialogue function. This natural language processing AI includes not only a single artificial intelligence having at least a translation function and a dialogue function, but also a case in which an AI with a translation function and an AI with a dialogue function work together and operate in an integrated manner. Furthermore, it is particularly preferable that the natural language processing AI used in the patent document translation method of this embodiment also has a grammar check function in addition to the translation function and the dialogue function. Furthermore, the natural language processing AI used in the patent document translation method of this embodiment may work in an integrated manner in cooperation with an image generation AI that has the function of generating and understanding various images (image data) such as drawings, tables, chemical formulas, and mathematical formulas.

[0024] Furthermore, in this specification, the term "dialogue function" refers to a user interacting with a natural language processing AI in the same way as humans converse or discuss freely, or to a user providing (inputting) instructions, commands, or requests, or various information, such as the document to be translated, to the natural language processing AI in any format and content. Therefore, the term "dialogue function" does not include a rigid interaction in which the AI ​​presents the user with pre-set, standardized options (e.g., options for the first language and the second language) set in a menu format and the user selects one option from the multiple options presented. The natural language processing AI used in the patent document translation method of this embodiment has the dialogue function defined above in addition to the translation function. Therefore, when setting the translation conditions in the translation condition assignment step described below, it is extremely easy for the user to easily and intuitively set the desired translation conditions, even without any specialized IT knowledge, such as programming languages. Furthermore, compared to a function-specific AI specialized only in translation, a natural language processing AI equipped with at least translation and dialogue functions offers extremely high flexibility and freedom in customizing the conditions for machine translation. In this regard, natural language processing AI, which has at least translation and dialogue functions, is extremely suitable for translating patent documents, which require many considerations during translation.

[0025] The natural language processing AI exists within a terminal computer 10 shown in FIG. 1 or on a network system 20 connected to the terminal computer 10. The terminal computer 10 includes a processing control unit 12 (e.g., a CPU), an operation unit 14 (e.g., a keyboard, a mouse, a stylus pen, a touch panel) connected to the processing control unit 12, a display unit 16 (e.g., a display), and a memory unit 18 (e.g., RAM, SSD, HDD), and may be connected to a network system 20 (e.g., the Internet, a local area network) via the processing control unit 12. In this case, the natural language processing AI can be used (i) as a cloud computing service provided on the network system 20, or (ii) as an AI application program installed on the memory unit 18.

[0026] In the case of (i) above, a user accesses a cloud computing service, or in the case of (ii) above, launches an AI application program, thereby enabling the user to utilize the natural language processing AI via the user interface displayed on the display unit 16 and the operation unit 14. As illustrated in FIG. 2 , the user interface 30 includes at least a text box (input field) 32, which is an area where the user inputs various information to the natural language processing AI, and a message display area 34, which displays various information transmitted from the natural language processing AI to the user (e.g., responses and messages to commands, questions, etc. sent by the user). The user interface 30 also preferably includes a navigation bar displaying a dialogue history, setting options, a button to start a new dialogue, and the like. The user can then utilize the translation and dialogue functions of the natural language processing AI via the text box 32 and message display area 34 that constitute the user interface 30 to implement the patent document translation method of this embodiment.

[0027] In the patent document translation method of this embodiment, the natural language processing AI that has at least the available translation function and dialogue function is not particularly limited, but examples include ChatGPT (provided by OpenAI), Google Assistant (provided by Google LLC), Microsoft Cortana (provided by Microsoft Corporation), IBM Watson Assistant (provided by IBM), and Baidu DuerOS (provided by Baidu, Inc.).

[0028] <<Machine Translation Process>> The patent document translation method of this embodiment is a patent document translation method in which the "specification" is machine-translated into a second language using an original text including the "claims" and "specification" written in a first language and a translation of the "claims" obtained by translating the "claims" into the second language.

[0029] The original text includes at least the "claims" and "description" of a patent document consisting of documents for application in a second country written in a first language (e.g., patent application documents written in a first language filed in a first country or documents with amendments thereto). The original text may also include an "abstract." If the "claims," ​​"description," and "abstract" contain image data (e.g., image data consisting of tables, chemical formulas, mathematical formulas, etc.), in principle, only the text data portion excluding these image data is used as the original text. However, in the patent document translation method of this embodiment, if an image generation AI is integrated in conjunction with a natural language processing AI, the "claims," ​​"description," and "abstract" may contain image data, and the original text may further include "drawings." The "claims," ​​"description," "abstract," and "drawings" included in the original text may be written not only in a clearly identifiable form as item titles, but also in a substantially identifiable form.

[0030] Furthermore, the translation of the claims into a second language is often a manual translation. Here, manual translation refers to (i) a human translation of the claims into a second language, or (ii) a machine translation of the claims, which is then checked and / or corrected by a human. Manual translation is typically preferably performed by a human (e.g., a translator) with appropriate translation skills. Therefore, a machine translation of the claims, which involves no substantial human intervention, is not used. Because claims are subject to review and are important for determining the scope of rights after granting a patent, translations must be accurate and legally compliant. Additionally, claims often contain claims (especially independent claims) that are long and complex, making them prone to translation errors when machine-translated. For these reasons, it is often inappropriate to simply use a machine-translated text as the translation of the claims.

[0031] The patent document translation method of this embodiment includes at least a reference text assignment step, a translation condition assignment step, and a translation step, and may further include other steps such as an error candidate extraction step as necessary. Each step will be described in detail below.

[0032] (Reference text assignment step) In the reference text assignment step, the natural language processing AI is assigned the translated "Claims" and the original text as reference text when performing machine translation. Specifically, the translated "Claims" and the original text are entered into text box 32 of user interface 30. At this time, it is preferable to further enter information that the translated "Claims" is a translation of the "Claims" contained in the original text into a second language. Furthermore, in the translation condition assignment step described below, from the perspective of facilitating the input of concise and clear translation conditions, it is also preferable to enter names assigned to the translated "Claims," ​​the original text, the "Claims," ​​and the "Description" so that they can be distinguished. For example, the names "Translation of the Claims," ​​"Original Text," "Claims," ​​and "Specification" can be assigned to the respective text information of the "Claims" translation, the original text, the "Claims," ​​and the "Specification," or the names "Text 2 CL," "Text 1," "Text 1 CL," and "Text 1 SP" can be assigned to them. Note that as the "Claims" translation and the original text to be used in the reference text assignment step, electronic data (such as text data) in a form that can be input into a natural language processing AI is prepared in advance.

[0033] (Translation condition setting step) In the translation condition assignment step, translation conditions for translating the "specification" into a second language are assigned to the natural language processing AI. Here, the translation conditions refer to conditions that specify how a document written in a first language should be translated into a second language. Specifically, the translation condition assignment step is performed by inputting the translation conditions into the text box 32 of the user interface 30. At a minimum, the translation conditions can be set to translate the "specification" provided to the natural language processing AI in the reference text assignment step into a second language by referring to the "claims" and the translation of the "claims" already provided to the natural language processing AI in the reference text assignment step. In this case, it is particularly preferable to set the translation conditions to translate the "specification" into a second language so that the translation of the "claims" is consistent with the translation of the "claims" in terms of wording. Setting such translation conditions and performing machine translation can significantly reduce the post-editing work required for the resulting translation of the "specification."

[0034] Meanwhile, when translating a document, there are two main types of translation styles: literal translation and free translation. Literal translation translates sentences and expressions into another language while preserving the original's words and structure, and also reflecting the original's word order and expressions as closely as possible. Free translation, on the other hand, preserves the meaning and message of the original text while paraphrasing it to reflect the context and expression of the original text. Compared to literal translation, this translation style emphasizes translating in a way that is easier for the reader to understand. When translating patent documents, not only is it necessary to accurately translate the content of the original, but when filing a second patent application claiming priority based on a first patent application, it is also generally required to translate as literal as possible in order to ensure the benefits of the priority right.

[0035] However, when it comes to machine translation using natural language processing AI, depending on the algorithm, the default translation style (the setting when simply performing machine translation without setting any particular translation conditions) can vary, such as being a literal translation or translation that leans more towards literal translation, or being a free translation or translation that leans more towards free translation, or being based on a literal translation but using free translation in cases where a literal translation would not preserve the meaning of the original text, so literal translation is not always performed.

[0036] In light of these circumstances, it is particularly preferable to set the translation conditions to employ a literal translation as the primary translation style, and specific examples of setting translation conditions in this case include (i) setting a literal translation of the entire "specification" when translating it into a second language, or (ii) setting a literal translation of the entire "specification" when translating it into a second language, but using a free translation only when a literal translation would not preserve the meaning of the original. Setting such translation conditions and performing machine translation can significantly reduce the post-editing work of the resulting translated "specification."

[0037] Furthermore, in neural machine translation, when a specific word written in a first language is translated into a second language, it is known that the specific word written in the first language may be translated by associating it with multiple translations written in the second language (see, for example, JP 2022-57482 A, paragraph 0005). Thus, in neural machine translation, the word written in the first language and the word translated into the second language may not be translated one-to-one. To reliably prevent such translation errors, it is also preferable to set a translation condition such that a specific word written in the first language is translated by assigning only one translation written in the second language. Setting such translation conditions and performing machine translation can significantly reduce the post-editing work of the resulting "specification" translation. Furthermore, in the error candidate extraction step described below, it becomes easier to more accurately extract words that are likely to be errors.

[0038] As explained above, it is preferable to set at least one of the following conditions (a) to (c) as the translation conditions, and it is more preferable to set two or more conditions in combination. (a) When translating the "specification" into a second language, the translation should conform to the wording of the translation of the "claims." (b) When translating the "Specification" into a second language, the "Specification" shall be translated word for word. (c) Each word in the text of the "specification" must be translated into a second language using only one translation.

[0039] The conditions (a) to (c) below are basic conditions, and therefore they can be modified to be more restrictive while maintaining the essence of the basic conditions. For example, condition (b) can be modified to be more restrictive, as in condition (b') below. Therefore, the above conditions (a) to (c) conceptually include conditions obtained by modifying these basic conditions in a more restrictive manner. (b') When translating the "Specification" into a second language, the "Specification" should be translated word for word in principle. However, if a word for word translation would not preserve the meaning of the original text, a free translation should be used.

[0040] (Translation step) In the translation step, the natural language processing AI generates a translation of the "specification" by translating the "specification" into a second language in accordance with the translation conditions set in the translation condition setting step. Specifically, after inputting instructions to perform translation and the translation conditions into text box 32 of user interface 30, the translation step can be performed by pressing the Enter key on the keyboard constituting operation unit 14 or by pressing an execute button displayed on user interface 30.

[0041] In the patent document translation method of this embodiment, the original text, including the "claims" and "description" written in the first language, is not directly machine-translated. Instead, a separate manual translation of the "claims" is prepared as reference material for machine translation, and the "description" is then machine-translated. Considering this, compared to simply machine-translating the entire source text, the patent document translation method of this embodiment has the disadvantage of complicating the translation process prior to machine translation. However, the patent document translation method of this embodiment still offers significant advantages, more than offsetting this disadvantage. The reasons for this are explained below.

[0042] First, among the various documents that can be translated, patent documents are unique in that they often contain lengthy and complex claims. Furthermore, as mentioned above, such lengthy and complex claims are prone to various translation errors due to the principles of neural machine translation. Therefore, simply translating the claims can lead to a significant burden of proofreading after the translation. Furthermore, consistency between the translated claims and the translated description is crucial to satisfying the support requirements. However, as mentioned above, when the claims translation obtained through machine translation contains various translation errors, proofreading after the translation is necessary not only to address each individual translation error but also to restore the consistency between the claims and the description before the translation. However, restoring consistency requires reviewing the entire source text before and after the translation, which makes the proofreading process extremely burdensome.

[0043] In contrast, in the patent document translation method of this embodiment, machine translation of the "specification" is performed using a previously properly translated "claims" translation as a reference material during machine translation. Therefore, there is no need to consider translation errors in the translated "claims" from the beginning, and consistency between the translated "claims" and the translated "specification" is likely to be maintained at approximately the same level as the original text. Therefore, when considering the entire translation process before and after machine translation, the patent document translation method of this embodiment makes it easier to create an accurate translation more efficiently than simply machine translating the entire original text, including the "claims" and the "specification."

[0044] In light of the above, when the first language is Japanese, the patent document translation method of this embodiment preferably requires that the number of characters in at least one claim included in the "claims" written in the first language be 300 or more, more preferably 450 or more, and even more preferably 600 or more. When the number of characters in at least one claim included in the "claims" written in Japanese is 300 or more, simply machine-translating the "claims" tends to result in more frequent translation errors. Therefore, in such cases, using the patent document translation method of this embodiment makes it easier to create an accurate translation more efficiently than simply machine-translating the entire original text. While there is no particular upper limit on the number of characters in at least one claim included in the "claims" written in the first language, it is generally preferable that the number be 3,000 or less, and more preferably 2,500 or less.

[0045] (Error candidate extraction step) After the translation step is completed, a further error candidate extraction step may be performed as necessary. In the error candidate extraction step, the natural language processing AI compares specific wording in the translated "Claims" with the entire translated "Description" to extract words from the translated "Description" that are not exactly identical to the specific wording but are highly similar to it. For example, in the case where the second language is English, if the specific wording in the translated "Claims" is "organic-inorganic particle" and the translated "Description" contains words such as "inorganic-organic particle" and "organic-inorganic fine particle" that are not identical to "organic-inorganic particle" but are highly similar to it, these words are extracted as error candidates. The criteria for determining the degree of similarity may simply be "a phrase that has a high degree of similarity in its wording compared to a specific phrase," but in order to improve the accuracy of the determination, a more specific determination criterion may be set, such as "a phrase that has a high degree of similarity compared to a specific phrase if it differs only in the order of the words that make up the phrase," in the case where the specific phrase is a combination of two or more words. Furthermore, the criteria for determining the degree of similarity are set on a case-by-case basis depending on the type of second language.

[0046] There are no particular restrictions on the form in which candidate error words are extracted, but for example, candidate error words in the text data of the translated "specification" can be underlined, highlighted in bold, or italicized, etc. Furthermore, if the translated "specification" includes a paragraph number such as

[0123] , candidate error words can be extracted by displaying a report in the message display area 34 of the user interface 30 that combines the paragraph number containing the candidate error word with the candidate error word.

[0047] This allows the user to refer to the extracted error candidates and appropriately correct the translation of the "specification" as necessary. For example, if "organic-inorganic particle" and "organic-inorganic fine particle" are simply variations in wording in the translation of the "specification," the user can determine that "organic-inorganic fine particle" is a typographical error. Also, if "organic-inorganic particle" and "organic-inorganic fine particle" are used in the translation of the "specification" as words with different technical meanings, the user can determine that "organic-inorganic fine particle" is not a typographical error.

[0048] As described above, in the patent document translation method of this embodiment, each step can be executed by a natural language processing AI. When executing each step, a user can provide various instructions, such as translation conditions, to the natural language processing AI in a natural dialogue format, similar to a human conversation or discussion. This has the advantage that a user can instruct the natural language processing AI with the desired instructions without any specialized IT knowledge, such as programming languages. However, on the other hand, the instructions may be ambiguous or unclear to the natural language processing AI, resulting in an inability to obtain the desired translation result. Even in such cases, however, the user can appropriately revise the instructions to obtain the desired translation result, taking into account the relationship between the specified instructions and the translation result obtained based on them. Repeating this feedback process can further improve the translation result desired by the user. This feedback process is also effective for scenario setting when setting up robotic process automation, as described below.

[0049] <<First and Second Languages>> In the patent document translation method of this embodiment, the first and second languages ​​may be the official languages ​​of each country. Typically, the first language can be the official language of the country in which the applicant resides, and the second language can be the official language of the foreign country in which the applicant wishes to obtain rights. However, it is preferable that the first language be one language selected from the group consisting of Japanese, English, Chinese, and Korean, and the second language be a language selected from the group consisting of Japanese, English, Chinese, and Korean that is different from the first language. Among these, a particularly preferable combination of the first language and the second language is when the first language is Japanese and the second language is English, or when the first language is English and the second language is Japanese.

[0050] <<Automating the machine translation process>> In the patent document translation method of this embodiment, the entire machine translation process, consisting of each step, may be performed through interaction between a user and a natural language processing AI via a user interface 30. Alternatively, part or all of the machine translation process may be performed using robotic process automation (RPA). This allows for the automation of part or all of the machine translation process. For example, if the entire machine translation process is automated using RPA, the user simply prepares the source text and the translation of the "claims" required in the reference text assignment step and provides them to the natural language processing AI to automatically obtain the translation of the "description." In this case, the user essentially only needs to arrange and prepare the source text and translation of the "claims" (pre-edited as necessary) and post-edit the translation of the "claims."

[0051] To enable natural language processing AI to operate using RPA, you simply need to set it up using the steps 1 to 5 below. 1. Integration: Obtain the API (Application Programming Interface) of the natural language processing AI and link the natural language processing AI and RPA via the API. 2. Scenario setting: Design a dialogue scenario between the user and natural language processing AI (providing (input) the original text, setting the translation conditions, etc.), and set the appropriate response from the natural language processing AI to the user's instructions. 3. Workflow construction: The RPA tool receives input from the natural language processing AI and sets the corresponding automated task (such as saving the text data of the translated "specification" in the memory unit 18 or a specified folder on the network system 20). 4. Error handling: We will establish a system to detect and deal with errors in the integration of natural language processing AI and RPA. 5. Test and production operation: Test the entire flow and make sure there are no problems before running it in the production environment.

[0052] <Example> Next, a specific example of the patent document translation method of this embodiment will be described. First, the information shown in " " below is entered into text box 32 of user interface 30, and the reference text assignment step is executed by pressing the Enter key on the keyboard constituting operation unit 14 or by clicking the execute button displayed on user interface 30 by clicking a button on the mouse constituting operation unit 14. As a result, a message indicating that the natural language processing AI has understood the input content is displayed in message display area 34.

[0053] Please read and understand the text in " " below: (1) Claims, (2) Specification, and (3) English Translation of Claims. (2) Specification is a document that explains (1) Claims in more detail, and (3) English Translation of Claims is an English translation of (1) Claims. (1) Scope of Claims “[Claim 1] A curable resin composition comprising a polymerizable monomer, a polymerization initiator, and organic-inorganic composite particles. [Claim 2] ..." (2) Specification "[Title of invention] Curable resin composition [Technical field]

[0001] The present invention relates to a curable resin composition. [Background technology]

[0002] ..." (3) English translation of the claims "CLAIMS [Claim 1] A curable resin composition comprising: a polymerizable monomer, a polymerization initiator, and organic-inorganic composite particles. [Claim 2] ..."

[0054] Next, the user enters the information shown in " " below into text box 32 of user interface 30, and executes the translation condition assignment step and translation step by pressing the Enter key on the keyboard that constitutes operation unit 14 or by clicking the execute button displayed on user interface 30 by clicking a button on the mouse that constitutes operation unit 14. As a result, (2) the English translation of the specification (machine translation) is displayed in message display area 34. The user can then obtain the post-edited English translation of the specification by appropriately post-editing (2) the English translation of the specification (machine translation) displayed in message display area 34.

[0055] "Please translate (2) the specification into English based on (1) the claims, (2) the specification, and (3) the English translation of the claims. However, please translate the specification in accordance with the following conditions (a) to (c)." (a)(2) When translating the specification into English, the translation should be in accordance with the wording of the English translation of the claims. (b)(2) Translate the entire text of the specification verbatim (no paraphrases). (c)(2) Each word in the text of the specification must be translated into English using only one translation.

[0056] It is also possible to further perform an error candidate extraction step on the English translation (machine-translated text) of the specification (2) displayed in the message display area 34 as a result of the execution of the translation condition assignment step and the translation step. In this case, the error candidate extraction step is executed by entering the information shown in " " below into the text box 32 of the user interface 30, and then pressing the Enter key on the keyboard constituting the operation unit 14 or clicking the execute button displayed on the user interface 30 by clicking a button on the mouse constituting the operation unit 14. As a result, the English translation (machine-translated text) of the specification (2) is displayed in the message display area 34 with wording that is a candidate for error displayed in bold italics. The user can then obtain the post-edited English translation of the specification by appropriately performing post-editing on the English translation (machine-translated text) of the specification (2) displayed in the message display area 34.

[0057] "In the English translation of the specification (2) shown above, please indicate in bold italics any wording that you believe to be a clerical error. Please determine whether or not there is a clerical error in accordance with the following conditions (a) to (c). (a)(3) When the wording contained in the English translation of the claims is similar to the wording contained in the English translation of the specification (2) (excluding cases where the two wordings are completely identical). In particular, when the following condition (b) or (c) is met, it must be determined that the two are similar. (b)(3) When the wording contained in the English translation of the claim consists of a single word, the wording that is presumed to be misspelled should be deemed similar. (c)(3) When the wording contained in the English translation of the claims is composed of a combination of two or more words, if the order of the words is different or if the number of words constituting the claim is one more or one less, it should be judged as similar. [Explanation of symbols]

[0058] 10: Terminal computer 12: Processing control section 14:Operation section 16: Display section 18: Storage section 20: Network Systems 30: User Interface 32: Text box 34: Message display area

Claims

1. A patent document translation method for machine-translating a "specification" into a second language using an original text including "claims" and a "description" written in a first language and a translation of the "claims" obtained by translating the "claims" into a second language, a reference text providing step of providing the translation of the Claims and the original text as reference texts when performing machine translation to a natural language processing artificial intelligence (NLP) that exists in a terminal computer or on a network system connected to the terminal computer and has at least a translation function and a dialogue function; a translation condition assigning step of assigning translation conditions to the natural language processing artificial intelligence for translating the "specification" into a second language; a translation step in which the natural language processing artificial intelligence translates the "specification" into a second language in accordance with the translation conditions to generate a translation of the "specification"; How to translate patent documents, including:

2. 2. The patent document translation method according to claim 1, further comprising a step of extracting candidate errors, in which the natural language processing artificial intelligence compares specific wording in the translated "Claims" with the entire translated "Specification" to extract wording from the translated "Specification" that is not completely identical to the specific wording but is highly similar to the specific wording.

3. 2. The patent document translation method according to claim 1, wherein the translation conditions include at least one of the following conditions (a) to (c): (a) When translating the "Description" into a second language, the translation should conform to the wording of the translation of the "Claims." (b) When translating the "Description" into a second language, the "Description" shall be translated word for word. (c) Each word in the text of the "specification" must be translated into a second language using only one translation.

4. 4. The patent document translation method according to claim 1, wherein the first language is any one language selected from the group consisting of Japanese, English, Chinese, and Korean, and the second language is selected from the group consisting of Japanese, English, Chinese, and Korean, and is a language different from the first language.

5. 4. The patent document translation method according to claim 1, wherein the first language is Japanese and the second language is English.

6. 4. The method for translating a patent document according to claim 1, wherein the first language is English and the second language is Japanese.

7. the first language is Japanese, A method for translating a patent document according to any one of claims 1 to 3, wherein the number of characters in at least one claim included in the "Claims" written in the first language is 300 characters or more.

8. The patent document translation method according to any one of claims 1 to 3, wherein the natural language processing artificial intelligence is any one selected from the group consisting of ChatGPT, Google Assistant, Microsoft Cortana, IBM Watson Assistant, and Baidu DuerOS.

9. The method for translating patent documents according to any one of claims 1 to 3, wherein part or all of the machine translation process is carried out using robotic process automation.

Images